Nightmare Fuel for Intel: Arm CEO Predicts Arm will Take Over 50% Windows PC Market-share by 2029

[londiste]

Can you explain the failure of this instruction: AVX-512? It turned out to be a power hog, and was abandoned.
Why was that? And what solution does RISC-V have to compute as an alternative?

AVX-512 is not an instruction, it is an instruction set extension. This does not directly affect the core x86-64.

It is a power hog mainly because it deals with 512-bit operands. Longer registers, wider execution units. There are good use cases for AVX-512, the hardware implementation just was not there and the question became whether it is worth it. For these instructions to work the hardware implementation can be different. AMD started supporting AVX-512 in Zen4 by using two 256-bit floating pipes together and only in Zen5 by extending the length of floating point units.

What do you mean what solution does RISC-V have to compute as an alternative?

AVX-512 does not come from empty space. There was MMX, then there were versions of SSE and 3DNow!, then AVX, AVX2 and now AVX-512. Width of operations increased along with that as well. Aside from operations themselves also the amount of data being processed. IIRC SSE brought 128-bit, AVX2 256-bit (and heat issues in Haswell and Skylake), AVX-512 brought 512-bit. There are use cases - mostly related to multimedia - where SIMD and wider data processing is beneficial.

ARM isn't different either. There were SIMD, VFP (not SIMD but vector), Advanced SIMD/NEON extensions at one time or another. These are the successful ones that survived (or a version for them did). And these were not a standard functionality in ARM until some point (v8?). This type of operations are now being added to ARM proper (at a somewhat slower pace).

RISC-V will have to basically walk the same path once it starts to work with the same or similar data or algorithms. Wider operations, bigger units. At some point they cannot affort to keep on doing things by 64-bit chunks when competition does 2, 4 or 8 times wider (for example see what AMD did with AVX2 and AVX-512 support - initially made do with smaller units and used 2 of them to do the operation but then made the units bigger in the next iteration). Whether that will end up with more complex operations is an interesting topic and we will have to see. In theory they can keep the hardware simple but that will only move the same complexity to software.

 

[Assimilator]

Only legacy x86 apps apps are keeping people on the ecosystem and they're slowing disappearing.

Legacy x86 apps like videogames, you mean?

 

[Daven]

This TPU April Fools story is looking a lot more likely.

Intel Realizes the Only Way to Save x86 is to Democratize it, Reopens x86 IP Licensing | TechPowerUp}

While 50% is a pull-from-his-ass kind of number, ARM laptop, desktop and server market share will definitely go up from what it is today. Intel can’t afford that. Fabs cost about the same money whether they are making chips or not and every 1% shift of market share from Intel to ARM drains Intel’s coffers faster and faster.

Intel will never get the kinds of third party orders its needs to keep those fabs at capacity as long as they make their own chips. Intel needs to pivot to a hybrid ARM-TSMC business model. They need to become a one stop shop for licensing IP and then fabbing the chips or split into two companies.

 

[azrael]

Have we actually seen any real tests of the new Snapdragon X processors? I only recall seeing different showcases.

Also, the success of Windows on ARM is very much dependent on two things, developer support and whether or not the ARM processors can run x86/x64 software satisfactorily in emulation. Most people don't want to give up their existing software.

 

[kondamin]

Have we actually seen any real tests of the new Snapdragon X processors? I only recall seeing different showcases.

Also, the success of Windows on ARM is very much dependent on two things, developer support and whether or not the ARM processors can run x86/x64 software satisfactorily in emulation. Most people don't want to give up their existing software.

Those reviews should be landing this week as embargo's should be lifted I think tomorrow or tonight.
Pretty certain results will be very good for general use and software that has gotten optimizations.

I think the bigger issue is going to come later when more chips enter the market that don't have all the extensions on soc or different configurations and make arm on windows a minefield where davinci resolve on a laptop with a snapdragon X works like a champ but on a laptop with snapdragon Y doesn't work at all as it's a .x hardwave version and blackmagic can't be bothered to fix it like they never bothered fixing for vega

 

[bitsandboots]

I've lost track of where Intel is going with regards to increasing competition and threats to their business (arm, amd, nvidia)
AMD's plan is more interesting to me. They're executing very well on x86 plans, but have yet to make a difference in the AI gold rush and so far don't seem to be gearing up to directly compete with arm.

Those reviews should be landing this week as embargo's should be lifted I think tomorrow or tonight.
Pretty certain results will be very good for general use and software that has gotten optimizations.

I think the bigger issue is going to come later when more chips enter the market that don't have all the extensions on soc or different configurations and make arm on windows a minefield where davinci resolve on a laptop with a snapdragon X works like a champ but on a laptop with snapdragon Y doesn't work at all as it's a .x hardwave version and blackmagic can't be bothered to fix it like they never bothered fixing for vega

Also predicting DOA for GPU activities that stray outside a narrow golden path. I have only ever heard negative things about qualcomm's adreno drivers relative to intel, amd, nvidia.
It's one reason the Apple chips aren't usable in some situations - because you can't reasonably pair the apple chips with a 3rd party gpu.
I hope to see the snapdragons be less of a walled garden. It'd be cool to see a qualcomm + nvidia pairing for example.

 

[Daven]

AMD's plan is more interesting to me.

It'd be cool to see a qualcomm + nvidia pairing for example.

Looks like you are a DIY PC enthusiast on a DIY PC enthusiast website. The ARM trend does not apply to thee. It is for the 99.999% of the rest of the world who don’t care about AMD, Intel and Nvidia. They just want a computer as our world becomes more computerized.

 

[SL2]

Also no wonder Dell keeps downplaying the existence of AMD and goes with Qualcomm today, probably Nvidia tomorrow.

Dell is also the only one (AFAIK) who's ever released a Radeon RX 7900M laptop. I never saw a Dell Phoenix laptop, but I have seen with Hawk point.

To me it seems like they're changing.

 

[bitsandboots]

Looks like you are a DIY PC enthusiast on a DIY PC enthusiast website. The ARM trend does not apply to thee. It is for the 99.999% of the rest of the world who don’t care about AMD, Intel and Nvidia. They just want a computer as our world becomes more computerized.

You're not completely wrong, but getting ARM into places where GPUs are important is still valuable!
Clearly qualcomm's attempts are bottom-up. Go for the least demanding markets first. Sure.
But, this doesn't just leave out "DIY PC enthusiasts". This also leaves out workstations.
And, also, DIY PC enthusiasts have viral marketing influence. If we get cool ARM computers, we're sure to spread adoption elsewhere by spreading the word.
So, I'm looking forward to ARM adoption NOT being a lock-in to a terrible GPU, so that all workloads can be uplifted.
ARM doesnt need to be paired with a bad gpu, so it shouldn't be limited to it, that's all.

 

[Daven]

You're not completely wrong, but getting ARM into places where GPUs are important is still valuable!
Clearly qualcomm's attempts are bottom-up. Go for the least demanding markets first. Sure.
But, this doesn't just leave out "DIY PC enthusiasts". This also leaves out workstations.
And, also, DIY PC enthusiasts have viral marketing influence. If we get cool ARM computers, we're sure to spread adoption elsewhere by spreading the word.
So, I'm looking forward to ARM adoption NOT being a lock-in to a terrible GPU, so that all workloads can be uplifted.
ARM doesnt need to be paired with a bad gpu, so it shouldn't be limited to it, that's all.

As a diehard DIY PC enthusiast, I want what you want. But workloads and storage are moving to the cloud, for the majority of the planet, a smartphone is the only ‘computer’ most will ever own and once an industry goes super mainstream, boutique parts are relegated to hobbiests.

In the future, discrete GPUs for desktops, consumer purchaseable socketed CPUs and other DIY parts will become antiques as the computing world moves to wearables, ‘dumb terminals’ that fit in your pocket and massive super computing arrays.

Having to sit in a stationary position (desk) while interacting with a piece of plastic (mouse) to use a box connected with wires (desktop) will look barbaric to users in the not so distant future.

 

[SL2]

Take some most popular ARM platform, that is arguably raspberryPI... can you boot a PCIe video card? It it will take years, decades for something to budge there.

Yes. It's not ready for consumers, but it works.

Snapdragon X even has eight unused PCIE 4 lanes.

 

[Ferrum Master]

Yes. It's not ready for consumers, but it works.

Snapdragon X even has eight unused PCIE 4 lanes.

You call it works out of the box, the memory mapping is a nightmare and most features are broken. That's the real life perspective of ARM using as a daily driver... Looking at Geerlings eyes at the end he had put few nighters with few hundred kernel compile tries and begging for code patches.

 

[SL2]

You call it works out of the box, the memory mapping is a nightmare and most features are broken.

Where did I say out of the box? What do you think not ready for consumers means???

You didn't get my point. If a end user could get that far with existing parts, it wouldn't take years for an OEM to develop something.

 

[Ferrum Master]

it wouldn't take years for an OEM to develop something.

For most OEMs there is no other source to even make the compiles for most devices. You expect charity from FOSS? You will get small fraction, and who will maintain it? No one. Same plague Android has... NO DRIVER, just because of the way people do business. We cannot have proper DRM driver because of Hollywood on X86 Radeon... what ARM?

It will take decades to have some sort of compatibility and make it a tool not a toy.

 

[SL2]

For most OEMs there is no other source to even make the compiles for most devices. You expect charity from FOSS? You will get small fraction, and who will maintain it? No one. Same plague Android has... NO DRIVER, just because of the way people do business. We cannot have proper DRM driver because of Hollywood on X86 Radeon... what ARM?

It will take decades to have some sort of compatibility and make it a tool not a toy.

I don't expect shit. Sorry if I was stepping on your toes.

Have a good day.

 

[john_]

Not gonna happen.

It would be almost impossible for Intel to return from this. The fabs will go to someone else, someone American or controllable from US government, as for the CPU maker part of the company, they will need to adjust to the new reality, as AMD did when they sold their fabs. And Intel's only hope will be TSMC in the first few years. But with ARM camp on the offensive and Intel not having an advanced node to build it's x86 CPUs is vast quantities, because TSMC will be making chips mostly for Nvidia, Apple and Qualcomm, that will be the end of x86. Intel might survive as a name, but it will not be the company you probably have in mind when posting "not gonna happen".

NVIDIA has no interest in wasting effort on specialised products for "gamers and enthusiasts" when they can and are making money hand over fist from putting more and more powerful GPUs in datacentres.

You are looking too much in today, but things change and Nvidia has proven many times that they think a few years ahead of others.
Nvidia tried to get an x86 license about 20(?) years ago. Intel said no. Then Intel killed third party chipsets for it's platform, one of those companies affected by that decision was Nvidia, having to shut down it's chipset department. But both those examples, Nvidia trying to get an x86 license and getting in the chipset business, shows that they where never intended to limit themselves in GPUs. Yes now it enjoys tremendous AI success, but why ignore the client market? Their marketing department talked about Premium AI PCs, meaning PCs using Nvidia GPUs. You think they wouldn't want to be selling ARM based and fully Nvidia made platforms to OEMs and not just GPUs? They wouldn't want to be the backbone of the majority of laptops, mini PCs and desktops in a few years, the same way Intel was all that time? AI market could be saturated in a few years with multiple solutions from multiple companies or just more hardware out there than needed. Why not have a big share of the client market in the future? Intel still makes billions from the client market. Nvidia could make more.

It's easy when you have a handful of applications and don't need to support games.

Of course it is. But that's also the reason why smartphones succeded. Because in the end the average consumer only uses a handful of applications. AAA gaming will follow, especially if Nvidia gets involved.

 

[ARF]

Nvidia tried to get an x86 license about 20(?) years ago. Intel said no.

That's not true. Since other companies have the x86 license.

However, it’s not the case. We see some company releasing x86 compatible chip like Zhaoxin. But most of these chips saw the light of day after the FTC forced Intel to allow other companies owning x86 IP to undergo mergers and joint ventures. Hence disseminating, under specific conditions, the capabilities to build x86 compatible chips.

The never-ending story of X86 patents expiration

Every couple of years, this story re-emerges, and everybody starts to talk about the upcoming expiry of x86 related patents. They were followed by several stories revolving emulation and the forthcoming flood of third-party implementations of compatible processors. But this never happens. Why?

www.blopeur.com

List of x86 manufacturers - Wikipedia

en.wikipedia.org

 

[Darmok N Jalad]

Hmm, where have I heard predictions like this before...

Microsoft Exec: People Are Choosing Windows Phones Over iPhone, Android Smartphones | CRN

Microsoft, nearly three years after launching Windows Phone, can now say it has overtaken a rival smartphone vendor. Now it's rallying partners around the news.

www.crn.com

Windows Phone will overtake iPhone by 2015, analyst divines

Analyst firm IDC has stuck its neck out with a futuregazing report that sees Windows Phone overhauling Apple's iOS in just four years.

www.cnet.com

 

[Onasi]

That's not true. Since other companies have the x86 license.

Erm, what’s not true? NVidia tried to get ahold of the x86 license, they even petitioned FTC for one and even tried to get one out of Intel during their settlement OF said FTC petition (and aforementioned chipset BS). They got a lot, including some cash, but explicitly NOT the x86 license and not even a permission to build an emulator for it.

 

[ARF]

Erm, what’s not true? NVidia tried to get ahold of the x86 license, they even petitioned FTC for one and even tried to get one out of Intel during their settlement OF said FTC petition (and aforementioned chipset BS). They got a lot, including some cash, but explicitly NOT the x86 license and not even a permission to build an emulator for it.

Settlements with bribes means that nvidia gave up on asking, otherwise it would have won the class-action lawsuit.

 

[john_]

That's not true.

If you don't know, then don't reply in such a way.
Nvidia had plans for an X86 CPU and those plans where out in the open. They tried to get a license from Intel, Intel said no. They tried to buy VIA, it didn't worked. Then Intel decided to not give license to third parties to build chipsets for Nehalem, which was the reason for Nvidia to close it's chipset division. The settlement between Intel and Nvidia was meant a couple of billions to Nvidia for damages because of that division closure, but at the same time Nvidia was forgetting even the idea of building an X86 CPU. They where in fact even agreeing to not even build an X86 emulator for their ARM based SOCs.

 

[Onasi]

Settlements with bribes means that nvidia gave up on asking, otherwise it would have won the class-action lawsuit.

That wasn’t a class-action, for one. And that was back in 2011, NVidia of that time was not the undisputed juggernaut it is today, so no, the idea that they “just lost interest” and otherwise would have been able to pressure Intel in court into handing over the x86 license has no basis in reality.

 

[londiste]

That's not true. Since other companies have the x86 license.

The never-ending story of X86 patents expiration

Every couple of years, this story re-emerges, and everybody starts to talk about the upcoming expiry of x86 related patents. They were followed by several stories revolving emulation and the forthcoming flood of third-party implementations of compatible processors. But this never happens. Why?

www.blopeur.com

From your own link:

Intel has also used patent lawsuits over the past 30 years to attempt to limit competition from Cyrix, Via, NVidia, and Transmeta, among others. Though they did lose or unfavourably settle a number of those lawsuits.

Nvidia trying to get an x86 license was a pretty big story back then. They did fish for some stranger ways but none worked out.
Currently the only companies with x86 license in addition to Intel and AMD should be VIA and whoever was behind the Vortex86.

Also, as the story on your own link said - patent on x86 itself might be expired but it isn't expired on extensions, some of which are critical/required for an x86 CPU today. Most notably x86-64 - which is technically AMD's and covered by the extensive mess of cross-licensing agreements among the x86 licensees.

 

[Denver]

That sounds to me more like improving the ARM marketshare currently close to irrelevance by 50%, and not magically capturing 50% of the market (that would be dumb to say).

 

[Daven]

That sounds to me more like improving the ARM marketshare currently close to irrelevance by 50%, and not magically capturing 50% of the market (that would be dumb to say).

This is a good take either way. Having the products that would in theory satisfy 50% of the needs of laptop users is the first step. This was not the case up until a few years ago as smartphone ARM SoCs were not up to the task. Build it and they will come (that's what Apple did). However, it always comes down to the voluntary choice of the customers which in the end is unpredictable so we will see how this plays out once we have a few years of laptop-class ARM SoCs available from Qualcomm, Mediatek, Nvidia, Apple and others.